Computer scientist Louis Castricato quit his doctoral studies at Brown University and founded Overworld, a company building AI that can understand and navigate a physical world rather than just processing words. He is part of a growing shift among prominent AI researchers who see chatbots as a diminishing frontier for fundamental innovation.
"We basically have passed the point of doing real fundamental LLM research," Castricato told Fast Company. "Now it's just applications." The move reflects a broader pivot away from large language models—the technology behind ChatGPT and Claude—toward so-called "world models" that teach AI systems how to react in real-world environments, sometimes through robots.
The shift comes even as investors continue committing trillions of dollars to leading chatbot developers like Anthropic and OpenAI. But a growing contingent of entrepreneurs and scientists argue that the next breakthroughs lie in embodied intelligence, not in further scaling language models.
Among the most prominent advocates is Fei-Fei Li, often called the "Godmother of AI," who described the concept of a world model as "one of the most important and most overloaded terms in AI today." The field seeks to give AI spatial reasoning, physical intuition, and the ability to interact with three-dimensional spaces.
Castricato's company Overworld signals a new wave of startups betting that the most valuable AI research will come from teaching systems to navigate the physical world. However, critics caution that world models remain largely experimental and lack the clear commercial playbook that made chatbots an instant multibillion-dollar industry.